Information, informality and trade: evidence from small traders in East Africa

Traders in the treatment group will receive access to a platform (through their phone) that provides trade information (market prices, exchange rates, taxes/tariffs and trade procedures).

2021-04-28

2022-06-30

1. Business outcomes: sales, profits, costs (including transport, tariffs/taxes, ... ), probability of trading/being in business, purchasing price of goods (including inv. hyp. sine or log transformations for all monetary variables when applicable)

2. Trade outcomes: type of goods traded (take-up of new goods), number of goods traded, quantity, number of supplying and selling markets, frequency/number of trips, value of goods per trip

3. Cross border trade: traders' choice of supplier/selling market (whether it is domestic versus cross border), probability of exchanging money

4. Informality and Border costs: traders’ choice of formal versus informal border crossing, probability of paying a bribe, level of the bribe, probability of facing harassment, other non-tariff barriers (also run on cross border sample)

5. Market level outcomes: market prices, market level trade flows, market competition measures / market structure

- For each family of outcomes outlined above, I will create an index. I will run two type of analysis, using the index created as well as run regressions separately for each outcome.
- Type 1 error will be controlled across the tested hypotheses
- I will show results for the full sample; however, I will winsorize continuous outcomes at 1 and 99 percentile.
- I consider market level outcomes (family 5) and trader level outcomes (family 1 - 4) as two distinct studies and allocate 5% type 1 error to each.

- Heterogeneity analysis of main outcomes by gender, trader size, trader type (cross border vs. domestic or formal vs. informal vs. domestic), industry, trader type x industry, age/experience of trader, education, prior beliefs of border costs (collected pre intervention)
- Border crossing experience outcomes, e.g., time day chosen for crossing, waiting time, negotiation experience
- Take up of platform (extensive margin) and frequency of use (intensive margin). In addition, type of information requested to platform through phone: number of requests, type of information requested
- Household wellbeing including food security
- Spillover of treatment at market level (using intensity of treatment design)
- Outcomes on value chain/suppliers: e.g., supplier's sales and profits, supplier's prices
- Effect of reminder SMS on main adoption/use of platform for traders
- For main outcomes, I will also look at ToT effect in addition to the ITT

- For each family of outcome listed in secondary outcomes end points, I will create an index. I will run two type of analysis, using the index created as well as run regressions separately for each outcome.
- I will show results for the full sample; however, I will winsorize continuous outcomes at 1 and 99 percentile.

A listing of traders will be carried out in markets located close to Busia (Uganda-Kenya crossing). A random sample of traders will be selected based on pre-selected stratas. A baseline will be administered to the sample of selected traders and their main supplier. A random sample of traders (treatment group) will receive access to a platform where they can access market prices, exchange rate information and information about tax rates and border crossing procedures. The control group will not receive access to this platform. High frequency data will be collected on trading behaviors on both treatment and control groups throughout the study. Finally, an endline study will be administered both on traders and main suppliers.

Analysis:

For each of the outcomes of interest, I will use the following specifications to estimate the effect of information for trader i in survey round t.

Primary analysis:
(i) I will analyze primary outcomes using the following specification (a) using all survey rounds including survey rounds pre implementation:
outcome_it = a + b_1 Treat_it + gamma_i + gamma_t + e_it
with gamma_i being a trader fixed effect and gamma_t a survey round fixed effect. Both fixed effects simply improve the precision with which I can estimate the coefficient of interest but are not necessary for causal identification.

Secondary analysis:

(ii) I will also run primary outcomes using a specification (b) using all survey rounds, of the type:

outcome_it = a + b_1 Treat_it + X_i + market_i + gamma_t + e_it

with X including trader characteristics collected as baseline such as gender, age/experience, industry, size/scale, trader type, nationality... and market_i being either market fixed effects or a control for market level intensity of treatment/saturation (trader's market at baseline/midline).

(iii) I will analyze secondary outcomes using specification (a) and (b) outlined above.

(iv) I will also run all primary outcomes and secondary outcomes for each survey round post implementation separately to show dynamic changes in estimated treatment effects as a function of time. The specification (c) for each round will be of the type:
outcome_i = a + b_1 Treat_i + delta X_i + market_i + e_i
with X including trader characteristics collected as baseline such as gender, age/experience, industry, size/scale, trader type, nationality, education... and market_i being either market fixed effects or a control for market level intensity of treatment/saturation (trader's market at baseline/midline).
Similarly, a specification (d) of the following type can be run on the whole panel:
outcome_it = a + b_1 Treat_i + b_2 Treat_i x Round_t + gamma_t + delta X_i + market_i + e_it
with Treat_i x Round_t being interaction variable between treatment and survey round.

Note 1: for all specifications mentioned above, I can replace survey round fixed effects by an indicator Post_it which equals 1 if the data was collected after implementation. An additional variable will then be added to control for border closure.
Note 2: for all specifications mentioned above, I can limit the survey rounds to those close (in time) to the implementation period, i.e., limit the analysis to baseline/midline and rounds post implementation.

Randomization done by a computer. The randomization is stratified by market, trader type (domestic, formal, informal) and gender.

Trader (individual) level randomization. In addition, I add variation in treatment intensity at market x industry level to account for potential spillovers.

Yes

For the analysis that controls for spillovers, I use the following clusters: ~40 markets (located close to the border) and 2 types of industry (agriculture and shoes/clothing). For the main specification, the randomization is however at the trader level.

1100 traders (agriculture and shoes/clothing), up to 1000-2000 suppliers (farmers or traders)

~500 treatment traders, 500 control traders